It sounds like the Credit Card Processing system at the cashier had internet for processing credit cards etc. but the waiter app has no internet dependencies since it can transfer the order to the cashier system.
Salary inflation is much trickier to measure because it is confounded by years of experience increasing, getting promotions, etc. The distribution of the working population by seniority also changes over time so it's not self-correcting across the distribution. Assuming you have a good way of measuring it, (salary inflation/price inflation) would be an interesting Financial Quality of Life measure.
He predicts it might be possible from model math but doesn't actually say what his prediction is. He also argues it's possible we are on a s-curve that levels out before superhuman intelligence.
How about the silly treadmill where we waste billions of compute to compute useless proof of work type behaviours and whenever more compute gets thrown at the problem we just make it harder to ensure there isn't better output. I believe it was called buttcoin or something silly like that.
Parts of this do align well with how to maximize profits. Shared walls, progressively smaller units over time and removing balconies have been the story of condo buildings over the last generation. The area that doesn’t line up is the low end apartment fixtures. It turns out people will pay $15k for $10k better of appliances and countertops.
This is a scapegoat at best in areas with high homelessness. The mentally ill and drug abusers are the most visible part of the homeless population not all the homeless population. There are also complex relations to cause vs effect where mental illness may be at manageable levels until a crisis like homelessness exacerbates it or drug use may be a result of being homeless instead of a cause.
The dilemma is you can’t make new housing cheap because if the price falls too much it becomes unprofitable to build and you don’t get new supply.
My city is currently facing this where the interest rate hikes, build tax hikes and falling prices have created a perfect storm of vastly reduced housing starts.
It's not clear to me it loses at all on point 4. It seems to me AI is quite far from producing human quality literature or even human quality news content for example so it's hard to see a direct line from OpenAI using book/news content to author's books/articles being worth less. Maybe you can argue AI summaries in search results reduces clicks to the New York Times or something like that but that would be more specific to search.
Point two is also a bit ambiguous because many of the works involved are weaker forms of copyright. Books for example don't generally have copyrighted violated if you write a summary or analysis of that book so if the AI model isn't reproducing content verbatim or only does so for limited purposes (like providing direct quotes) it is probably non-infringing from an output perspective.
Point one by itself isn't a bright line test as well as plenty of commercial uses have been found to be non-infringing.
Almost all work is tax deductible. This isn't just for software developers and employees still pay taxes. The issue is software development is being classified as research which is one of the few areas where salaries aren't directly deductible but instead need to be deducted in pieces over five years.
Because employee salaries are normally deductible as an expense for companies which generally pay taxes on something resembling profit rather than something resembling revenue. The issue is they are classifying software development as a research activity so the salary has to be amortized over five years with 20% of the expense applying in each of the five years.
AI companies aren't seriously arguing that copyright shouldn't apply to them because "it's bad for business". The main argument is that they qualify for fair use because their work is transformative which is one of the major criteria for fair use. Fair use is the same doctrine that allows a school to play a movie for educational purposes without acquiring a license for the public performance of that movie. The original works don't have model weights and can't answer questions or interact with a user so the output is substantially different from the input.
The challenge with fan out is constructing a linear conversation that makes sense that captures previous history. In any context where the LLM needs that information linear loops often perform better than trying to splice together conversations from multiple parallel processes.
LLMs give you superpowers if you work on a wide range of tasks and aren't a deep expert on all of them. They aren't that great if you know the area you are working in inside and out and never stray from it.
For example, using an LLM to help you write a Dockerfile when you write Dockerfiles once a project and don't have a dedicated expert like a Deployment Engineer in your company is fantastic.
Or using an LLM to get answers faster than google for syntax errors and other minor issues is nice.
Even using LLM with careful prompting to discuss architecture tradeoffs and get its analysis (but make the final decision yourself) can be helpful.
Generally, you want to be very careful about how you constrain the LLM through prompts to ensure you keep it on a very narrow path so that it doesn't do something stupid (as LLMs are prone to do), you also often have to iterate because LLMs will occasionally do things like hallucinate APIs that don't actually exist. But even with iteration it can often make you faster.
My CPAP does the first two. But generally if you get event the tiniest leak (and it's hard to stop all the leaks) it's very noisy for my partner. My masks leaks so little that I have between 0 and 1 events per hour but it's still annoyingly loud so my partner usually sleeps in our guest bedroom. If I could take a pill instead I would do so gladly. That being said my CPAP machine typically reduces events by 98% or so, so 56% just isn't going to cut it.
This is the country of free speech zones away from the main event in the early 2000s and sending WWI dissenters to jail in 1914. You’ve long pretended to have freedoms you’ve never actually been given and this is hardly new.
I still believe in having humans do PRs. It's far cheaper to have the judgement loop on the AI come before and during coding than after. My general process with AI is to explicitly instruct it not to write code, agree on a correct approach to a problem and if the project has any architectural components a correct architecture then once we've negotiated the correct way of doing things ask it to write code. Usually each step of this process takes multiple iterations of providing additional information or challenging incorrect assumptions of the AI. I can get it much faster than human coding with a similar quality bar assuming I iterate until a high quality solution is presented. In some cases the AI is not good enough and I fall back to human coding but for the most part I think it makes me a faster coder.
Fair use allows for certain uses of copyrighted works without a specific license for those works. One of the major criterion is how transformative the work is and an LLM model is very different from the original work so it seems likely that criterion at least is met.
I think the headline is a bit misleading. Mets did pirate the works but may be entitled to use them under fair use. It seems like the authors are setting up for failure by making the case about whether the AI generation hinders the market for books. AI book writing is such a tiny segment what these models do that if needed Meta would simply introduce guard rails to prevent copying the style of an author and continue to ingest the books. I also don’t think AI generated fiction is anywhere near high quality enough to substantially reduce the market for the original author.
I think it depends a lot on your finances though. If you come from a rich family and have parental support by all means it is amazing to travel young. But if your travel budget is coming out of your downpayment on your house that could easily be the difference between buying before house prices got out of control or not. For example if you could have bought in 2013 without travel and it takes you till 2015 to save up a 2013 downpayment but in 2015 house prices have gone up and your downpayment needs to be larger and it takes more time, etc.
The context is completely different though. Building a normal computer app is not an attempt to do anything without government or legal structures so it makes sense that normal computer apps would be protected by government or legal structures.
It doesn't really make sense for people to build smart contracts that are intended to be an extra-judical agreement where the code enforces the rules and then run to government whenever something they don't like happens. What is the purpose of smart contracts at all if you still need the entire legal apparatus around them?
What does agreeing to a contract that inherently implies trying to work around the need for government in contracts means? What does it say about intent?
If for example, the firm that lost money had been saying "Code is Law" in their previous pro-crypto statements and had explicitly talked about smart contracts being extra-judical it seeems there intent would be to avoid legal intervention entirely and it would require a fairly high bar to argue that any bug could result in a lawsuit.
My supervising professor for the PhD program I left did a paper on the Chinese Room and argued that to a large degree understanding of the task was the ability to compress it many orders of magnitude. In that sense the LLMs are succeeding because despite their supposively massive parameter sets they are absolutely tiny compared to the Chinese Room version.
After trying to set up wireless on my printer interface and enter a password with up and down arrows rotating through an entire set of keys, I'm fairly convinced that no wireless on the iPod was massively correct. If people were expected to set up wifi by entering a password with a rotation device adoption would be miniscule.
I think the illegal monopoly claims are a bit out there given the range of offerings in the social network space. Are they arguing that Instagram specifically is a monopoly in photo-oriented social networking because X is mostly text and YouTube and TikTok is mostly video? I don't see any particular time you can point to where Facebook+Instagram+WhatsApp was a monoopoly in any sort of broad social networking space, especially not the WhatsApp part which competes with iMessage which is an absolute behemoth in terms of market share.
On the other hand the top ten comments as a whole are 3 bullish, 5 neutral, 2 bearish which is certainly not an overwhelming sentiment in any direction. That's despite the fact that bullish comments on start-ups tend to get more votes because it's a start-up community.
Large aggregation is also to a certain degree a moat. Most creators have quickly found that people won't pay for one creator's content unless that creator is a huge volume creator (at the scale of maybe Disney). No one subscribes to a platform with 20 movies and 5 TV shows.
I think it's less about wanting to be called an artist and more about wanting to be able to get the results of making things more easily. And there is a good case that we can have a successful boom from this sort of thing.
I'd argue many of the clones were better but not enough better. Doom was the first to do X and once you already played a game that did X the next game needed to do 2X not 1.2X.
Styles are notoriously hard to copy. It's not like Ghibli style being copied is anything new. For example, Studio Ponoc where former Ghibli employees did the copying. As long as they aren't literally generating copyrighted works styles can't be owned.
Would the world be better off if Picasso's heirs owned Cubism and any artist wanting to produce works in that style had to buy a license?